Synthetic pretraining dataset FinePhrase tops 2.8M downloads on Hugging Face
yb2698 · x · 2026-11-17
Joel Niklaus announced that FinePhrase, a synthetic pretraining dataset, has surpassed 2.8 million downloads on Hugging Face.
Built from FineWeb-Edu, the dataset uses SmolLM2-1.7B-Instruct to rewrite documents into four formats—FAQs, math problems, tables, and tutorials—totaling 1 billion rows, generated with DataTrove and speculative decoding. All four configs are open for download.
More from Research
- RL agent beats Pokémon Red with a <10M-parameter policy, 60,500x smaller than DeepSeek-V3 — Bam4d · 2026-11-17
- Verus Proof Index: new formal verification benchmark ranks 25 models, GPT-6 Astra leads at 59.1 — fhuszar · 2026-11-17
- Harmonic's AI theorem prover Aristotle takes a no-hallucination hardware synthesis talk to Cambridge — satnam6502 · 2026-11-17
- François Fleuret: why doesn't causal attention's asymmetric information flow hurt? — francoisfleuret · 2026-11-17
- Skild AI open-sources Steel physics solver: up to 13x faster than PhysX with 3x less GPU memory — deepakpathak · 2026-11-17
- MIT's Jacob Andreas trains models to explain their own computations — MacrinePhD · 2026-11-17